github/nohang
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix PSI support

Browse Source
This commit is contained in:
Alexey Avramov 2019-02-25 19:01:53 +09:00
parent 0c9d89ac5a
commit 7e34a6e03d
1 changed files with 56 additions and 81 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

133
nohang
View File

@ -101,18 +101,18 @@ def errprint(text):
def mlockall(): def mlockall():
MCL_CURRENT = 1 MCL_CURRENT = 1
MCL_FUTURE = 2 MCL_FUTURE = 2
MCL_ONFAULT = 4 MCL_ONFAULT = 4
libc = CDLL('libc.so.6', use_errno=True) libc = CDLL('libc.so.6', use_errno=True)
result = libc.mlockall( result = libc.mlockall(
MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT
) )
if result != 0: if result != 0:
result = libc.mlockall( result = libc.mlockall(
MCL_CURRENT | MCL_FUTURE MCL_CURRENT | MCL_FUTURE
) )
if result != 0: if result != 0:
print('Cannot lock all memory') print('Cannot lock all memory')
else: else:
@ -353,10 +353,6 @@ def pid_to_name(pid):
'utf-8', 'ignore').partition('\n')[0] 'utf-8', 'ignore').partition('\n')[0]
def pid_to_ppid(pid): def pid_to_ppid(pid):
try: try:
with open('/proc/' + pid + '/status') as f: with open('/proc/' + pid + '/status') as f:
@ -375,9 +371,6 @@ def pid_to_ppid(pid):
ppid = f_list[i].split('\t')[1] ppid = f_list[i].split('\t')[1]
def pid_to_ancestry(pid): def pid_to_ancestry(pid):
anc_list = [] anc_list = []
while True: while True:
@ -390,7 +383,6 @@ def pid_to_ancestry(pid):
print('Ancestry: ', anc_list) print('Ancestry: ', anc_list)
def pid_to_cmdline(pid): def pid_to_cmdline(pid):
""" """
Get process cmdline by pid. Get process cmdline by pid.
@ -477,7 +469,6 @@ def send_notify_warn():
) )
''' '''
''' '''
body = 'MemAvail: {}%\nSwapFree: {}%'.format( body = 'MemAvail: {}%\nSwapFree: {}%'.format(
round(mem_available / mem_total * 100), round(mem_available / mem_total * 100),
@ -495,7 +486,7 @@ def send_notify_warn():
notify_helper_path, notify_helper_path,
round(mem_available / mem_total * 100), round(mem_available / mem_total * 100),
round(swap_free / (swap_total + 0.1) * 100) round(swap_free / (swap_total + 0.1) * 100)
) )
t0 = time() t0 = time()
os.system(b) os.system(b)
@ -519,8 +510,8 @@ def send_notify(signal, name, pid):
# symbol '&' can break notifications in some themes, # symbol '&' can break notifications in some themes,
# therefore it is replaced by '*' # therefore it is replaced by '*'
'&', '*' '&', '*'
)
) )
)
if root: if root:
# send notification to all active users with notify-send # send notification to all active users with notify-send
notify_helper(title, body) notify_helper(title, body)
@ -542,7 +533,7 @@ def send_notify_etc(pid, name, command):
name.replace('&', '*'), name.replace('&', '*'),
pid, pid,
command.replace('&', '*') command.replace('&', '*')
) )
if root: if root:
# send notification to all active users with notify-send # send notification to all active users with notify-send
notify_helper(title, body) notify_helper(title, body)
@ -580,6 +571,7 @@ def get_pid_list():
pid_list = get_pid_list() pid_list = get_pid_list()
def get_non_decimal_pids(): def get_non_decimal_pids():
non_decimal_list = [] non_decimal_list = []
for pid in pid_list: for pid in pid_list:
@ -612,12 +604,10 @@ def fattest():
pid_badness_list = [] pid_badness_list = []
if print_proc_table: if print_proc_table:
print(' PID badness Name eUID') print(' PID badness Name eUID')
print('------- ------- --------------- ----------') print('------- ------- --------------- ----------')
for pid in pid_list: for pid in pid_list:
# find and modify badness (if it needs) # find and modify badness (if it needs)
@ -654,7 +644,7 @@ def fattest():
str(badness).rjust(7), str(badness).rjust(7),
pid_to_name(pid).ljust(15), pid_to_name(pid).ljust(15),
pid_to_uid(pid).rjust(10) pid_to_uid(pid).rjust(10)
) )
) )
except FileNotFoundError: except FileNotFoundError:
@ -664,12 +654,12 @@ def fattest():
pid_badness_list.append((pid, badness)) pid_badness_list.append((pid, badness))
# Make list of (pid, badness) tuples, sorted by 'badness' values # Make list of (pid, badness) tuples, sorted by 'badness' values
#print(pid_badness_list) # print(pid_badness_list)
pid_tuple_list = sorted( pid_tuple_list = sorted(
pid_badness_list, pid_badness_list,
key=itemgetter(1), key=itemgetter(1),
reverse=True reverse=True
)[0] )[0]
pid = pid_tuple_list[0] pid = pid_tuple_list[0]
@ -682,8 +672,8 @@ def fattest():
pid_to_name(pid), pid_to_name(pid),
victim_badness, victim_badness,
round((time() - ft1) * 1000) round((time() - ft1) * 1000)
)
) )
)
return pid, victim_badness return pid, victim_badness
@ -715,7 +705,6 @@ def find_victim_and_send_signal(signal):
if n is ppid_index: if n is ppid_index:
ppid = line.split('\t')[1] ppid = line.split('\t')[1]
if n is uid_index: if n is uid_index:
uid = line.split('\t')[2] uid = line.split('\t')[2]
continue continue
@ -780,7 +769,6 @@ def find_victim_and_send_signal(signal):
if i is uid_index: if i is uid_index:
uid = f_list[i].split('\t')[2] uid = f_list[i].split('\t')[2]
if i is vm_size_index: if i is vm_size_index:
vm_size = kib_to_mib( vm_size = kib_to_mib(
int(f_list[i].split('\t')[1][:-3])) int(f_list[i].split('\t')[1][:-3]))
@ -887,7 +875,6 @@ def find_victim_and_send_signal(signal):
realpath, realpath,
cmdline) cmdline)
if execute_the_command and signal is SIGTERM and name in etc_dict: if execute_the_command and signal is SIGTERM and name in etc_dict:
command = etc_dict[name] command = etc_dict[name]
@ -925,7 +912,6 @@ def find_victim_and_send_signal(signal):
try: try:
m = check_mem_and_swap() m = check_mem_and_swap()
ma = round(int(m[0]) / 1024.0) ma = round(int(m[0]) / 1024.0)
sf = round(int(m[2]) / 1024.0) sf = round(int(m[2]) / 1024.0)
@ -945,16 +931,13 @@ def find_victim_and_send_signal(signal):
key = 'Send {} to {}'.format( key = 'Send {} to {}'.format(
sig_dict[signal], name) sig_dict[signal], name)
if signal is SIGKILL and post_kill_exe != '': if signal is SIGKILL and post_kill_exe != '':
os.system( os.system(
post_kill_exe.replace( post_kill_exe.replace(
'$PID', pid).replace( '$PID', pid).replace(
'$NAME', pid_to_name(pid) '$NAME', pid_to_name(pid)
)
) )
)
if gui_notifications: if gui_notifications:
send_notify(signal, name, pid) send_notify(signal, name, pid)
@ -1038,10 +1021,10 @@ def sleep_after_check_mem():
print( print(
'Sleep time: {} sec; (t_mem={}, t_swap={}, t_zram={})'.format( 'Sleep time: {} sec; (t_mem={}, t_swap={}, t_zram={})'.format(
round(t, 2), round(t, 2),
round(t_mem, 2), round(t_mem, 2),
round(t_swap, 2), round(t_swap, 2),
round(t_zram, 2) round(t_zram, 2)
) )
) )
@ -1217,7 +1200,7 @@ try:
etc_command = a[1].strip() etc_command = a[1].strip()
if len(etc_name) > 15: if len(etc_name) > 15:
errprint('Invalid config, the length of the process ' errprint('Invalid config, the length of the process '
'name must not exceed 15 characters\nExit') 'name must not exceed 15 characters\nExit')
exit(1) exit(1)
etc_dict[etc_name] = etc_command etc_dict[etc_name] = etc_command
@ -1253,7 +1236,6 @@ except FileNotFoundError:
exit(1) exit(1)
# print(processname_re_list) # print(processname_re_list)
# print(cmdline_re_list) # print(cmdline_re_list)
# print(uid_re_list) # print(uid_re_list)
@ -1650,14 +1632,12 @@ stdout.flush()
sigterm_psi = sigterm_psi_avg10 sigterm_psi = sigterm_psi_avg10
sigkill_psi = sigkill_psi_avg10 sigkill_psi = sigkill_psi_avg10
psi_min_sleep_time_after_action = psi_avg10_sleep_time
########################################################################## ##########################################################################
if psi_support and not ignore_psi: if psi_support and not ignore_psi:
kill_psi_t0 = time() + psi_avg10_sleep_time psi_t0 = time() + psi_avg10_sleep_time
term_psi_t0 = time() + psi_avg10_sleep_time
avg_value = '' avg_value = ''
@ -1681,22 +1661,22 @@ while True:
if print_mem_check_results: if print_mem_check_results:
avg_value = 'PSI mem some avg10: {} | '.format(str(avg10).rjust(6)) avg_value = 'PSI mem some avg10: {} | '.format(str(avg10).rjust(6))
if avg10 >= sigkill_psi and time() - kill_psi_t0 >= psi_min_sleep_time_after_action: if avg10 >= sigkill_psi and time() - psi_t0 >= psi_avg10_sleep_time:
time0 = time() time0 = time()
mem_info = 'avg ({}) > sigkill_psi ({})'.format(avg10, sigkill_psi) mem_info = 'avg ({}) > sigkill_psi ({})'.format(avg10, sigkill_psi)
find_victim_and_send_signal(SIGKILL) find_victim_and_send_signal(SIGKILL)
kill_psi_t0 = time() psi_t0 = time()
elif avg10 >= sigterm_psi and time() - term_psi_t0 >= psi_min_sleep_time_after_action: continue
if avg10 >= sigterm_psi and time() - psi_t0 >= psi_avg10_sleep_time:
time0 = time() time0 = time()
mem_info = 'avg ({}) > sigterm_psi ({})'.format(avg10, sigterm_psi) mem_info = 'avg ({}) > sigterm_psi ({})'.format(avg10, sigterm_psi)
find_victim_and_send_signal(SIGTERM) find_victim_and_send_signal(SIGTERM)
term_psi_t0 = time() psi_t0 = time()
else: continue
pass
mem_available, swap_total, swap_free = check_mem_and_swap() mem_available, swap_total, swap_free = check_mem_and_swap()
# print(mem_available, swap_total, swap_free) # print(mem_available, swap_total, swap_free)
# если метры - получаем киб выше и сразу. см. # если метры - получаем киб выше и сразу. см.
@ -1741,9 +1721,9 @@ while True:
if mem_report: if mem_report:
speed = delta / 1024.0 / report_delta speed = delta / 1024.0 / report_delta
speed_info = ' | ΔMem, M/s: {}'.format( speed_info = ' | ΔMem: {} M/s'.format(
str(round(speed)).rjust(5) str(round(speed)).rjust(5)
) )
# Calculate 'swap-column' width # Calculate 'swap-column' width
swap_len = len(str(round(swap_total / 1024.0))) swap_len = len(str(round(swap_total / 1024.0)))
@ -1755,7 +1735,7 @@ while True:
human(mem_available, mem_len), human(mem_available, mem_len),
just_percent_mem(mem_available / mem_total), just_percent_mem(mem_available / mem_total),
speed_info speed_info
) )
) )
elif swap_total > 0 and mem_used_zram == 0: elif swap_total > 0 and mem_used_zram == 0:
@ -1766,21 +1746,21 @@ while True:
human(swap_free, swap_len), human(swap_free, swap_len),
just_percent_swap(swap_free / (swap_total + 0.1)), just_percent_swap(swap_free / (swap_total + 0.1)),
speed_info) speed_info)
) )
else: else:
print('{}MemAvail: {} M, {} % | SwapFree: {} M, {} % | Mem' print('{}MemAvail: {} M, {} % | SwapFree: {} M, {} % | Mem'
'UsedZram: {} M, {} %{}'.format( 'UsedZram: {} M, {} %{}'.format(
avg_value, avg_value,
human(mem_available, mem_len), human(mem_available, mem_len),
just_percent_mem(mem_available / mem_total), just_percent_mem(mem_available / mem_total),
human(swap_free, swap_len), human(swap_free, swap_len),
just_percent_swap(swap_free / (swap_total + 0.1)), just_percent_swap(swap_free / (swap_total + 0.1)),
human(mem_used_zram, mem_len), human(mem_used_zram, mem_len),
just_percent_mem(mem_used_zram / mem_total), just_percent_mem(mem_used_zram / mem_total),
speed_info speed_info
) )
) )
# если swap_min_sigkill задан в абсолютной величине и Swap_total = 0 # если swap_min_sigkill задан в абсолютной величине и Swap_total = 0
if swap_total > swap_min_sigkill_kb: # If swap_min_sigkill is absolute if swap_total > swap_min_sigkill_kb: # If swap_min_sigkill is absolute
@ -1797,7 +1777,7 @@ while True:
# это для печати меминфо. Все переработать нахрен. # это для печати меминфо. Все переработать нахрен.
# далее пошло ветвление # далее пошла проверка превышения порогов
# MEM SWAP KILL # MEM SWAP KILL
if mem_available <= mem_min_sigkill_kb and \ if mem_available <= mem_min_sigkill_kb and \
@ -1820,11 +1800,11 @@ while True:
swap_sigkill_pc) swap_sigkill_pc)
find_victim_and_send_signal(SIGKILL) find_victim_and_send_signal(SIGKILL)
kill_psi_t0 = time() psi_t0 = time()
term_psi_t0 = time() continue
# ZRAM KILL # ZRAM KILL
elif mem_used_zram >= zram_max_sigkill_kb: if mem_used_zram >= zram_max_sigkill_kb:
time0 = time() time0 = time()
mem_info = '{}\nMemory statu' \ mem_info = '{}\nMemory statu' \
@ -1838,11 +1818,11 @@ while True:
percent(zram_max_sigkill_kb / mem_total)) percent(zram_max_sigkill_kb / mem_total))
find_victim_and_send_signal(SIGKILL) find_victim_and_send_signal(SIGKILL)
kill_psi_t0 = time() psi_t0 = time()
term_psi_t0 = time() continue
# MEM SWAP TERM # MEM SWAP TERM
elif mem_available <= mem_min_sigterm_kb and \ if mem_available <= mem_min_sigterm_kb and \
swap_free <= swap_min_sigterm_kb: swap_free <= swap_min_sigterm_kb:
time0 = time() time0 = time()
@ -1865,11 +1845,11 @@ while True:
swap_sigterm_pc) swap_sigterm_pc)
find_victim_and_send_signal(SIGTERM) find_victim_and_send_signal(SIGTERM)
kill_psi_t0 = time() psi_t0 = time()
term_psi_t0 = time() continue
# ZRAM TERM # ZRAM TERM
elif mem_used_zram >= zram_max_sigterm_kb: if mem_used_zram >= zram_max_sigterm_kb:
time0 = time() time0 = time()
mem_info = '{}\nMemory status that r' \ mem_info = '{}\nMemory status that r' \
@ -1885,12 +1865,11 @@ while True:
find_victim_and_send_signal(SIGTERM) find_victim_and_send_signal(SIGTERM)
# сделать одно время для обоих уровней. # сделать одно время для обоих уровней.
kill_psi_t0 = time() psi_t0 = time()
term_psi_t0 = time() continue
# -> psi_t0
# LOW MEMORY WARNINGS # LOW MEMORY WARNINGS
elif gui_low_memory_warnings: if gui_low_memory_warnings:
if mem_available <= mem_min_warnings_kb and \ if mem_available <= mem_min_warnings_kb and \
swap_free <= swap_min_warnings_kb + 0.1 or \ swap_free <= swap_min_warnings_kb + 0.1 or \
@ -1904,9 +1883,5 @@ while True:
print(time() - t0, 'send notify warning time') print(time() - t0, 'send notify warning time')
warn_timer = 0 warn_timer = 0
sleep_after_check_mem()
# SLEEP BETWEEN MEM CHECKS # SLEEP BETWEEN MEM CHECKS
else: sleep_after_check_mem()
sleep_after_check_mem()